Human Activity Recognition Using Accelerometer

GUIDE: R. Govardhan Reddy , ASSISTANT PROFESSOR

PROJECT BATCH NUMBER:CBIT/IT/2021-22/221
Hemanth Mallari (160118737097)
Jaswanth Nath Kuraganti (160118737099)
CONTENTS
INTRODUCTION
SENSORS LIKE ACCELEROMETER
PROBLEM STATEMENT
LITERATURE SURVEY
SOFTWARE AND HARDWARE REQUIREMENTS
SYSTEM FLOW DIAGRAM
CONCLUSION
BIBILOGRAPHY
INTRODUCTION
• Human activity Recognition is in massive demand due to its applications in Health care
domain, Computer Vision, Household safety and Robot Learning.
• A huge amount of money can be saved if sensors collect and monitor data of patients. System
can send automatically send reports to doctor in case of any abnormal behavior.
• Sensors of low cost like Accelerometer to identify human activities. Extensive growth in
popularity, accessibility and computations power of Accelerometer makes it ideal candidate
for non-invasive body attached sensor.
• Human Activity Recognition aims to identify the actions carried out by a person given a set of
observations of him/her and the surrounding environment.
SENSORS LIKE ACCELEROMETER
• An Accelerometer is a device which is used for measuring static or dynamic acceleration
forces.
• We can find out the angle by which the device is tilted by determining the static acceleration
due to gravity. and we can sense the direction of movement by measuring the dynamic
acceleration.
• Accelerometer may work on the piezoelectric effect or by sensing the changes in capacitance.
• Accelerometer contains seismic mass circuit which is made of silicon. The mass changes its
orientation according to the orientation of the device.
PROBLEM STATEMENT
• Human Activities from Accelerometer, predicting what a person is based on a trace of
their movements using these sensors.
• There is no clear analytical way to relate the sensor data to specific actions in a general
way.
• Activities such as walking, running, sitting , laying are classified as regular physical
movement is to be recognized in various applications like medical, computer vision and
robot learning.
LITERATURE SURVEY
Paper Title Author Name Method Advantages Disadvantages
Triaxial accelerometer- Wahyu Andhyka 1-dimensional • It can classify static For all such activities,
based human activity Kusuma, Agus Eko Convolution Neural and dynamic the precision values
recognition using 1D Minamo, Mochamad Network (CNN). activity using are varied between
convolution neural Satriyo Wibowo. dataset. low range(85%) for
network • Dynamic activity downstairs activity to
IEEE(2020) such as sitting high range(100%) for
achieves the Dynamic activity.
highest precision
value of 100%
Probabilistic Approach Walid Gomaa Probability • The approach has Can be complicated for
to Human Activity distribution, shown larger data sets, and
Recognition from Dissimilarity measure, effectiveness in the also loses the temporal
Accelerometer Data inertial Measurement predictive succession of the
IEEE(2020) unit. performance as streaming.
well as
computationally in
small data sets.
LITERATURE SURVEY
Paper Title Author Name Method Advantages Disadvantages
Classifying Human Kazuki Motani, LSTM Network Got an accuracy Only applicable to
Activity and Kelvin Wong, models are a type of with 83.2% accuracy some activities and
Smartphone Holding Shunsuke Kamijo recurrent neural of three activities. parameter
Mode Using network that are adjustment in each
Accelerometer and able to learn and mode.
Gyroscope. remember over long
IEEE(2019) sequences of input
data.
Human Activity Mst. Alema Khatun, Random Forest, K- When compared to When compared to
Recognition Using Mohammad Abu Nearest Neighbour machine learning, CNN, Random forest
Smartphone Sensor Yousuf. and Convolutional Deep learning has and Support vector
Based on Selective neural better results than machine has lesser
Classifiers network(CNN) Machine Learning. accuracy.
IEEE(2020) CNN algorithm gave RF- 73%
99% accuracy. KNN– 87%
LITERATURE SURVEY
Paper Title Author Name Method Advantages Disadvantages
Human Activity Ramano Luca, k-nearest neighbors, Smaller number of Only specific
Recognition using Silviu-loan Bejinariu, J48 and Random features is an movement of
Inertial Data Hariton Costin, forests classifiers. advantage. activities is
IEEE(2021) Florin Rotaru, Here, The features observed and
Gladiola Petriou. are extracted i.e classified , they are
acceleration of standing, walking
objects using Mean and going up-down
Decrease Impurity staird
(MDI)
SOFTWARE AND HARDWARE
REQUIREMENTS
Software requirements Hardware requirements

Operating System Windows 10 Processor Intel i5

Technology Open CV Hard disk 500 GB

IDE Jupyter/python IDE RAM 4GB/8GB

METHODOLOGY
CONCLUSION
The most suitable features required for classification of above-mentioned activities i.e Mean,
Median along with frequency domain features extracted from signal magnitude vector.

Past experiments have proved that triaxial accelerometer data can be used to classify these
type of human activities.

LOSO and 10-fold cross validation techniques of validate model using Random forest and KNN.

Future work could include including more finer human activities in the classification and
implement the algorithm in real-time system like smart phone.
BIBILOGRAPHY
IEEE paper – Sensing Technology of Large Dataset – Ong chin Ann and Lau Bee
theng.
IEEE paper – Human Activity Recognition System from different poses with CNN –
Md. Atikuzzaman, tarafdar Razibur Rahman, Eashitha wazed, Anishka Reddy.
IEEE paper - Human Activity Recognition using Recurrent Neural networks –
Deepika Singh, erinic merdivan, ismini Psychoula, Johannes Kropt.
S.S. Intille, F. Albinali, S. Mota, B. Kuris, P. Botana, W.L. Haskell. Design of a wearable
physical activity monitoring system using mobile phones and accelerometers.
Moeslund, A.Hilton,V.Kruger, “A survey of advances in vision-based human motion
capture and analysis, Computer Vision Image Understanding
THANK YOU